xref: /openbmc/linux/fs/ocfs2/buffer_head_io.c (revision e23feb16)
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * io.c
5  *
6  * Buffer cache handling
7  *
8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9  *
10  * This program is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU General Public
12  * License as published by the Free Software Foundation; either
13  * version 2 of the License, or (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public
21  * License along with this program; if not, write to the
22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23  * Boston, MA 021110-1307, USA.
24  */
25 
26 #include <linux/fs.h>
27 #include <linux/types.h>
28 #include <linux/highmem.h>
29 
30 #include <cluster/masklog.h>
31 
32 #include "ocfs2.h"
33 
34 #include "alloc.h"
35 #include "inode.h"
36 #include "journal.h"
37 #include "uptodate.h"
38 #include "buffer_head_io.h"
39 #include "ocfs2_trace.h"
40 
41 /*
42  * Bits on bh->b_state used by ocfs2.
43  *
44  * These MUST be after the JBD2 bits.  Hence, we use BH_JBDPrivateStart.
45  */
46 enum ocfs2_state_bits {
47 	BH_NeedsValidate = BH_JBDPrivateStart,
48 };
49 
50 /* Expand the magic b_state functions */
51 BUFFER_FNS(NeedsValidate, needs_validate);
52 
53 int ocfs2_write_block(struct ocfs2_super *osb, struct buffer_head *bh,
54 		      struct ocfs2_caching_info *ci)
55 {
56 	int ret = 0;
57 
58 	trace_ocfs2_write_block((unsigned long long)bh->b_blocknr, ci);
59 
60 	BUG_ON(bh->b_blocknr < OCFS2_SUPER_BLOCK_BLKNO);
61 	BUG_ON(buffer_jbd(bh));
62 
63 	/* No need to check for a soft readonly file system here. non
64 	 * journalled writes are only ever done on system files which
65 	 * can get modified during recovery even if read-only. */
66 	if (ocfs2_is_hard_readonly(osb)) {
67 		ret = -EROFS;
68 		mlog_errno(ret);
69 		goto out;
70 	}
71 
72 	ocfs2_metadata_cache_io_lock(ci);
73 
74 	lock_buffer(bh);
75 	set_buffer_uptodate(bh);
76 
77 	/* remove from dirty list before I/O. */
78 	clear_buffer_dirty(bh);
79 
80 	get_bh(bh); /* for end_buffer_write_sync() */
81 	bh->b_end_io = end_buffer_write_sync;
82 	submit_bh(WRITE, bh);
83 
84 	wait_on_buffer(bh);
85 
86 	if (buffer_uptodate(bh)) {
87 		ocfs2_set_buffer_uptodate(ci, bh);
88 	} else {
89 		/* We don't need to remove the clustered uptodate
90 		 * information for this bh as it's not marked locally
91 		 * uptodate. */
92 		ret = -EIO;
93 		put_bh(bh);
94 		mlog_errno(ret);
95 	}
96 
97 	ocfs2_metadata_cache_io_unlock(ci);
98 out:
99 	return ret;
100 }
101 
102 int ocfs2_read_blocks_sync(struct ocfs2_super *osb, u64 block,
103 			   unsigned int nr, struct buffer_head *bhs[])
104 {
105 	int status = 0;
106 	unsigned int i;
107 	struct buffer_head *bh;
108 
109 	trace_ocfs2_read_blocks_sync((unsigned long long)block, nr);
110 
111 	if (!nr)
112 		goto bail;
113 
114 	for (i = 0 ; i < nr ; i++) {
115 		if (bhs[i] == NULL) {
116 			bhs[i] = sb_getblk(osb->sb, block++);
117 			if (bhs[i] == NULL) {
118 				status = -EIO;
119 				mlog_errno(status);
120 				goto bail;
121 			}
122 		}
123 		bh = bhs[i];
124 
125 		if (buffer_jbd(bh)) {
126 			trace_ocfs2_read_blocks_sync_jbd(
127 					(unsigned long long)bh->b_blocknr);
128 			continue;
129 		}
130 
131 		if (buffer_dirty(bh)) {
132 			/* This should probably be a BUG, or
133 			 * at least return an error. */
134 			mlog(ML_ERROR,
135 			     "trying to sync read a dirty "
136 			     "buffer! (blocknr = %llu), skipping\n",
137 			     (unsigned long long)bh->b_blocknr);
138 			continue;
139 		}
140 
141 		lock_buffer(bh);
142 		if (buffer_jbd(bh)) {
143 			mlog(ML_ERROR,
144 			     "block %llu had the JBD bit set "
145 			     "while I was in lock_buffer!",
146 			     (unsigned long long)bh->b_blocknr);
147 			BUG();
148 		}
149 
150 		clear_buffer_uptodate(bh);
151 		get_bh(bh); /* for end_buffer_read_sync() */
152 		bh->b_end_io = end_buffer_read_sync;
153 		submit_bh(READ, bh);
154 	}
155 
156 	for (i = nr; i > 0; i--) {
157 		bh = bhs[i - 1];
158 
159 		/* No need to wait on the buffer if it's managed by JBD. */
160 		if (!buffer_jbd(bh))
161 			wait_on_buffer(bh);
162 
163 		if (!buffer_uptodate(bh)) {
164 			/* Status won't be cleared from here on out,
165 			 * so we can safely record this and loop back
166 			 * to cleanup the other buffers. */
167 			status = -EIO;
168 			put_bh(bh);
169 			bhs[i - 1] = NULL;
170 		}
171 	}
172 
173 bail:
174 	return status;
175 }
176 
177 int ocfs2_read_blocks(struct ocfs2_caching_info *ci, u64 block, int nr,
178 		      struct buffer_head *bhs[], int flags,
179 		      int (*validate)(struct super_block *sb,
180 				      struct buffer_head *bh))
181 {
182 	int status = 0;
183 	int i, ignore_cache = 0;
184 	struct buffer_head *bh;
185 	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
186 
187 	trace_ocfs2_read_blocks_begin(ci, (unsigned long long)block, nr, flags);
188 
189 	BUG_ON(!ci);
190 	BUG_ON((flags & OCFS2_BH_READAHEAD) &&
191 	       (flags & OCFS2_BH_IGNORE_CACHE));
192 
193 	if (bhs == NULL) {
194 		status = -EINVAL;
195 		mlog_errno(status);
196 		goto bail;
197 	}
198 
199 	if (nr < 0) {
200 		mlog(ML_ERROR, "asked to read %d blocks!\n", nr);
201 		status = -EINVAL;
202 		mlog_errno(status);
203 		goto bail;
204 	}
205 
206 	if (nr == 0) {
207 		status = 0;
208 		goto bail;
209 	}
210 
211 	ocfs2_metadata_cache_io_lock(ci);
212 	for (i = 0 ; i < nr ; i++) {
213 		if (bhs[i] == NULL) {
214 			bhs[i] = sb_getblk(sb, block++);
215 			if (bhs[i] == NULL) {
216 				ocfs2_metadata_cache_io_unlock(ci);
217 				status = -EIO;
218 				mlog_errno(status);
219 				goto bail;
220 			}
221 		}
222 		bh = bhs[i];
223 		ignore_cache = (flags & OCFS2_BH_IGNORE_CACHE);
224 
225 		/* There are three read-ahead cases here which we need to
226 		 * be concerned with. All three assume a buffer has
227 		 * previously been submitted with OCFS2_BH_READAHEAD
228 		 * and it hasn't yet completed I/O.
229 		 *
230 		 * 1) The current request is sync to disk. This rarely
231 		 *    happens these days, and never when performance
232 		 *    matters - the code can just wait on the buffer
233 		 *    lock and re-submit.
234 		 *
235 		 * 2) The current request is cached, but not
236 		 *    readahead. ocfs2_buffer_uptodate() will return
237 		 *    false anyway, so we'll wind up waiting on the
238 		 *    buffer lock to do I/O. We re-check the request
239 		 *    with after getting the lock to avoid a re-submit.
240 		 *
241 		 * 3) The current request is readahead (and so must
242 		 *    also be a caching one). We short circuit if the
243 		 *    buffer is locked (under I/O) and if it's in the
244 		 *    uptodate cache. The re-check from #2 catches the
245 		 *    case that the previous read-ahead completes just
246 		 *    before our is-it-in-flight check.
247 		 */
248 
249 		if (!ignore_cache && !ocfs2_buffer_uptodate(ci, bh)) {
250 			trace_ocfs2_read_blocks_from_disk(
251 			     (unsigned long long)bh->b_blocknr,
252 			     (unsigned long long)ocfs2_metadata_cache_owner(ci));
253 			/* We're using ignore_cache here to say
254 			 * "go to disk" */
255 			ignore_cache = 1;
256 		}
257 
258 		trace_ocfs2_read_blocks_bh((unsigned long long)bh->b_blocknr,
259 			ignore_cache, buffer_jbd(bh), buffer_dirty(bh));
260 
261 		if (buffer_jbd(bh)) {
262 			continue;
263 		}
264 
265 		if (ignore_cache) {
266 			if (buffer_dirty(bh)) {
267 				/* This should probably be a BUG, or
268 				 * at least return an error. */
269 				continue;
270 			}
271 
272 			/* A read-ahead request was made - if the
273 			 * buffer is already under read-ahead from a
274 			 * previously submitted request than we are
275 			 * done here. */
276 			if ((flags & OCFS2_BH_READAHEAD)
277 			    && ocfs2_buffer_read_ahead(ci, bh))
278 				continue;
279 
280 			lock_buffer(bh);
281 			if (buffer_jbd(bh)) {
282 #ifdef CATCH_BH_JBD_RACES
283 				mlog(ML_ERROR, "block %llu had the JBD bit set "
284 					       "while I was in lock_buffer!",
285 				     (unsigned long long)bh->b_blocknr);
286 				BUG();
287 #else
288 				unlock_buffer(bh);
289 				continue;
290 #endif
291 			}
292 
293 			/* Re-check ocfs2_buffer_uptodate() as a
294 			 * previously read-ahead buffer may have
295 			 * completed I/O while we were waiting for the
296 			 * buffer lock. */
297 			if (!(flags & OCFS2_BH_IGNORE_CACHE)
298 			    && !(flags & OCFS2_BH_READAHEAD)
299 			    && ocfs2_buffer_uptodate(ci, bh)) {
300 				unlock_buffer(bh);
301 				continue;
302 			}
303 
304 			clear_buffer_uptodate(bh);
305 			get_bh(bh); /* for end_buffer_read_sync() */
306 			if (validate)
307 				set_buffer_needs_validate(bh);
308 			bh->b_end_io = end_buffer_read_sync;
309 			submit_bh(READ, bh);
310 			continue;
311 		}
312 	}
313 
314 	status = 0;
315 
316 	for (i = (nr - 1); i >= 0; i--) {
317 		bh = bhs[i];
318 
319 		if (!(flags & OCFS2_BH_READAHEAD)) {
320 			/* We know this can't have changed as we hold the
321 			 * owner sem. Avoid doing any work on the bh if the
322 			 * journal has it. */
323 			if (!buffer_jbd(bh))
324 				wait_on_buffer(bh);
325 
326 			if (!buffer_uptodate(bh)) {
327 				/* Status won't be cleared from here on out,
328 				 * so we can safely record this and loop back
329 				 * to cleanup the other buffers. Don't need to
330 				 * remove the clustered uptodate information
331 				 * for this bh as it's not marked locally
332 				 * uptodate. */
333 				status = -EIO;
334 				put_bh(bh);
335 				bhs[i] = NULL;
336 				continue;
337 			}
338 
339 			if (buffer_needs_validate(bh)) {
340 				/* We never set NeedsValidate if the
341 				 * buffer was held by the journal, so
342 				 * that better not have changed */
343 				BUG_ON(buffer_jbd(bh));
344 				clear_buffer_needs_validate(bh);
345 				status = validate(sb, bh);
346 				if (status) {
347 					put_bh(bh);
348 					bhs[i] = NULL;
349 					continue;
350 				}
351 			}
352 		}
353 
354 		/* Always set the buffer in the cache, even if it was
355 		 * a forced read, or read-ahead which hasn't yet
356 		 * completed. */
357 		ocfs2_set_buffer_uptodate(ci, bh);
358 	}
359 	ocfs2_metadata_cache_io_unlock(ci);
360 
361 	trace_ocfs2_read_blocks_end((unsigned long long)block, nr,
362 				    flags, ignore_cache);
363 
364 bail:
365 
366 	return status;
367 }
368 
369 /* Check whether the blkno is the super block or one of the backups. */
370 static void ocfs2_check_super_or_backup(struct super_block *sb,
371 					sector_t blkno)
372 {
373 	int i;
374 	u64 backup_blkno;
375 
376 	if (blkno == OCFS2_SUPER_BLOCK_BLKNO)
377 		return;
378 
379 	for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
380 		backup_blkno = ocfs2_backup_super_blkno(sb, i);
381 		if (backup_blkno == blkno)
382 			return;
383 	}
384 
385 	BUG();
386 }
387 
388 /*
389  * Write super block and backups doesn't need to collaborate with journal,
390  * so we don't need to lock ip_io_mutex and ci doesn't need to bea passed
391  * into this function.
392  */
393 int ocfs2_write_super_or_backup(struct ocfs2_super *osb,
394 				struct buffer_head *bh)
395 {
396 	int ret = 0;
397 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
398 
399 	BUG_ON(buffer_jbd(bh));
400 	ocfs2_check_super_or_backup(osb->sb, bh->b_blocknr);
401 
402 	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) {
403 		ret = -EROFS;
404 		mlog_errno(ret);
405 		goto out;
406 	}
407 
408 	lock_buffer(bh);
409 	set_buffer_uptodate(bh);
410 
411 	/* remove from dirty list before I/O. */
412 	clear_buffer_dirty(bh);
413 
414 	get_bh(bh); /* for end_buffer_write_sync() */
415 	bh->b_end_io = end_buffer_write_sync;
416 	ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &di->i_check);
417 	submit_bh(WRITE, bh);
418 
419 	wait_on_buffer(bh);
420 
421 	if (!buffer_uptodate(bh)) {
422 		ret = -EIO;
423 		put_bh(bh);
424 		mlog_errno(ret);
425 	}
426 
427 out:
428 	return ret;
429 }
430